diff --git a/README.rst b/README.rst
index 46ccbc2..dc9b7da 100644
--- a/README.rst
+++ b/README.rst
@@ -4,7 +4,7 @@ PAULE
 .. image:: https://zenodo.org/badge/355606517.svg
    :target: https://zenodo.org/badge/latestdoi/355606517
 
-Predictive Articulatory speech synthesis Utilizing Lexical Embeddings (PAULE) a
+Predictive Articulatory speech synthesis Utilizing Lexical Embeddings (PAULE) is a
 python frame work to plan control parameter trajectories for the VocalTractLab
 simulator for a target acoustics or semantic embedding.
 
@@ -54,6 +54,9 @@ To cite the PAULE source code use the DOI 10.5281/zenodo.7252431
 (https://zenodo.org/doi/10.5281/zenodo.7252431), if you want to cite the
 software in general or the specific DOI on Zenodo.
 
+Linked Projects
+---------------
+To generate suitable corpus data  for the PAULE model, we have also implemented  `create_vtl_corpus <https://github.com/quantling/create_vtl_corpus>`_
 
 Acknowledgements
 ----------------
diff --git a/docs/source/index.rst b/docs/source/index.rst
index 9186f23..a278186 100644
--- a/docs/source/index.rst
+++ b/docs/source/index.rst
@@ -3,6 +3,10 @@
    You can adapt this file completely to your liking, but it should at least
    contain the root `toctree` directive.
 
+
+PAULE
+======
+Predictive Articulatory speech synthesis Utilizing Lexical Embeddings (PAULE) is a python frame work to plan control parameter trajectories for the VocalTractLab simulator for a target acoustics or semantic embedding.
 .. toctree::
    :maxdepth: 2
    :caption: Contents:
diff --git a/docs/source/training.rst b/docs/source/training.rst
new file mode 100644
index 0000000..dbb7f35
--- /dev/null
+++ b/docs/source/training.rst
@@ -0,0 +1,30 @@
+Training
+==========
+
+
+
+Data
+=====
+
+Paule requires data to be in the following format:
+
+
+
+
+Training
+========
+For effective training you probably want to use a GPU. 
+
+
+Paule consist of a number of models that are trained seperately. The models are:
+
+- `Embedder`  Input: Log mel spectrogram, Output: Semantic embedding, is added to the target embedding
+
+- `ForwardModel` Input: Normalized control parameters, Output: Log mel spectrogram
+
+- `InverseModel` Input: Log mel spectrogram, Output: Normalized control parameters
+
+-`MelGANs`
+
+
+
diff --git a/paule/paule.py b/paule/paule.py
index f6bcbc4..63c27d8 100644
--- a/paule/paule.py
+++ b/paule/paule.py
@@ -162,7 +162,7 @@ def __init__(self, *, pred_model=None, pred_optimizer=None, inv_model=None, inv_
         if embedder:
             self.embedder = embedder
         else:
-            self.embedder = EmbeddingModel(num_lstm_layers=2, hidden_size=720).double()
+            self.embedder = EmbeddingModel(num_lstm_layers=2, hidden_size=720, dropout=0.7).double()
             self.embedder.load_state_dict(torch.load(
                 os.path.join(DIR, "pretrained_models/embedder/embed_model_common_voice_syn_rec_2_720_0_dropout_07_noise_6e05_rmse_lr_00001_200.pt"),
                 map_location=self.device))